Reliably Accurate State Estimation for Connected and Autonomous Highway Vehicles

Farrell: Reliably Accurate State Estimation for Connected and Autonomous Highway Vehicles

September 27, 2019

Jay FarrellUC Riverside's Jay Farrell presented Reliably Accurate State Estimation for Connected and Autonomous Highway Vehicles on Sept. 27, 2019 at 4 p.m. in 290 Hearst Memorial Mining Building at the ITS Transportation Seminar.


Accurate and reliable awareness of world interactions is a key requirement for effective commercial deployment of autonomous and connected vehicles. Awareness arises from onboard sensors and ubiquitous communication between vehicles and infrastructure. Vehicle coordination and safety necessitate reliable “where-in-lane” knowledge of vehicle state. This presentation will address sensor fusion for high-bandwidth vehicle state estimation with a focus on high accuracy and reliability.

Advances is sensing and computation have dramatically altered the focus of related research. For example, computer vision and Global Navigation Satellite Systems each separately provide far more measurements than are necessary for observability. Such environments are signal-rich. The large number of measurements provides both opportunities (e.g., high accuracy) and challenges (e.g., large numbers of outliers). Standard state estimation approaches that decide irrevocably at each time which measurements are valid (e.g. EKF) are not sufficiently reliable at removing the effects of spurious measurements. When that decision is wrong, either measurement information is lost or the state and covariance estimates become corrupted, rendering all subsequent decisions suspect. Either situation can result in divergence of the state estimate, with potentially tragic consequences.

This presentation will consider moving horizon nonlinear state estimation by a novel risk-averse performance-specified (RAPS) approach. Moving horizon methods extract the Bayesian optimal trajectory using all sensor data over a temporal window (e.g. SLAM and RHE). RAPS modifies the optimization problem to select the least risky set of measurements that satisfies a user-defined performance constraint. RAPS is able to evaluate, and reconsider, outlier assumptions for all measurements within the temporal window. The presentation will include experimental results.

Jay A. Farrell earned B.S. degrees in physics and electrical engineering from Iowa State University, and M.S. and Ph.D. degrees in electrical engineering from the University of Notre Dame. While in the Autonomous Vehicles Group at Draper Lab, he received the Engineering Vice President's Best Technical Publication Award in 1990, and Recognition Awards for Outstanding Performance and Achievement in 1991 and 1993. He is a Professor in the Department of Electrical and Computer Engineering at the University of California, Riverside. He has served the IEEE Control Systems Society (CSS) on the Board of Governors for two terms, as Vice President Finance and Vice President of Technical Activities, as General Chair of IEEE CDC 2012, and as President in 2014. He has served on the Board of the Electrical and Computer Engineering Department Heads Association, as IEEE Education Activity Board treasurer, as a member of the IEEE Financial Committee, and three terms on the IEEE Fellow Committee. He currently serves as Vice President of the American Automatic Control Council. He was named a GNSS Leader to Watch for 2009-2010 by GPS World Magazine in May 2009 and a winner of the Connected Vehicle Technology Challenge by the U.S. Department of Transportation`s (DOT`s) Research and Innovative Technology Administration in July 2011. He is author of over 250 technical publications and three books, a Distinguished Member of IEEE CSS, a Fellow of AAAS, and a Fellow of the IEEE.